IDrive's RemotePC MCP Puts AI Assistants in Charge of Remote IT Tasks

IDrive launched RemotePC MCP on November 12, 2026, letting customers use AI assistants such as ChatGPT, Claude, Cursor, Codex and Grok to inspect and manage computers already configured in RemotePC. The hosted integration is available now, and it raises an immediate governance question for HR and IT leaders: how much remote-access authority should an AI agent hold, and for how long.

The launch touches people-process decisions directly. Remote support is a routine task for hybrid-work IT teams, managed service providers and remote employees, and the product documentation describes a single endpoint that covers every configured and online computer in an account. That removes the need for separate integrations or sign-ins per machine, a shift that changes how support staff structure their workflows rather than what they can reach. The feature does not enroll unmanaged devices, and it requires the regular RemotePC agent to be configured and online.

What can an AI assistant actually do?

The RemotePC FAQ separates observation from action. View Only sessions can retrieve registered devices and their status, account and billing details, groups, sub-users, recently accessed computers, visible application windows and screenshots. Full Control exposes action-based tools:

• Scheduling or canceling restarts, with a configurable grace period and optional user warning
• Checking for and installing RemotePC host updates
• Generating temporary connection links
• Displaying one-way messages on the remote screen

Practical examples from the company include a technician asking an assistant to inspect a screenshot and explain a visible application error, a remote employee checking whether an office computer is online, and an administrator warning a user before a planned restart. Defined tool calls return structured results that agents can chain into multi-step tasks, without requiring a full remote-control session for every routine check.

Raghava Meeniga, IDrive's vice president of technology, said the integration lets customers "make requests in natural language while retaining control over access permissions and session duration."

How does authorization work?

Customers authorize the connection with OAuth 2.1 and their existing RemotePC sign-in, including two-factor authentication when enabled. RemotePC states that sign-in credentials are not shared with the AI client. Every session carries an explicit View Only or Full Control permission and a duration between 12 hours and 30 days. When that period ends, authorization expires, and later tool calls fail rather than renewing silently.

Access can end early through My Account > MCP Connections, and revocation takes effect immediately, even during an active conversation, though it may interrupt operations in progress. One caution stands out for policy writers: disconnecting inside an AI client may not revoke the underlying authorization, so a closed conversation does not by itself prove that account access has ended.

What are the limits?

Each host handles one tool call at a time and returns a busy/retry response if another action is underway. Screenshots arrive through short-lived signed URLs rather than inline image data. Screen capture and window listing are blocked when a computer is locked, displaying a screen saver or showing a secure desktop. The service supports hosted MCP only, so customers cannot deploy their own server.

Setup differs by client. The ChatGPT guide uses a custom MCP plugin configured with OAuth; the Claude guide uses a custom connector and also documents command-line configuration. Both require an active RemotePC account and access to the relevant AI service, after which users sign in, choose the session timeout and permission level, and authorize access.

For HR and IT leaders drafting AI acceptable-use policies, the practical change is the interface, not the footprint. Host restrictions, explicit permission levels, expiring sessions and account-level revocation now determine how much authority an assistant receives, and organizations that define those settings centrally will be best positioned as AI-mediated device management spreads.